Effective features extraction and selection for hand gesture recognition using sEMG signal.

Surface Electromyographic (sEMG) signals are a promising approach to hand and finger gesture recognition. Most of the sEMG-based hand gesture recognition has developed based on the whole hand gesture, full wavelength, and all extracted features. However, further improvement of the recognition accuracy and reducing time complexity with effective feature extraction methods are still challenges for sEMG gesture recognition. Surface Electromyographic (sEMG) signals hold potential for hand and finger gesture recognition. While many sEMG-based hand gesture recognition methods rely on whole-hand gestures, full wavelength, and all extracted features, challenges remain in enhancing recognition accuracy, reducing time complexity, and effectively extracting features. In our study, we introduced a novel method for sEMG-based hand gesture recognition, emphasizing improving recognition accuracy and time efficiency. Our method integrates segmentation, effective feature extraction, and a potential feature selection approach to address these challenges. We captured the sEMG signal's significant motion from multiple channels using a sliding window and the MAD approach. From each channel, we extracted eighteen TD and FD features, yielding 144 features for the MA21 dataset and 360 for the UC8 dataset. We employed the LR algorithm for feature selection, enhancing our system's efficiency. Four ML classifiers, namely ETC, RF, SVM, and KNN, were tested on both segmented and full wavelength sEMG signals. The ETC outperformed, achieving peak accuracies of 97.33% (segmented) and 97.26% (full wavelength) for MA21 and 99.88% and 99.70% for UC8. Our model surpassed existing methods by over 5% in accuracy, highlighting its efficiency and superior capability. [ABSTRACT FROM AUTHOR]